Skip to content
This repository has been archived by the owner on Oct 8, 2020. It is now read-only.

spinnaker/keiko

Keiko

Keiko is a simple, at-least-once queueing library originally built for Spinnaker's Orca µservice.

Concepts

Keiko consists of a number of components:

The Queue

The queue itself accepts unique messages with an associated delivery time/ By default the delivery time is immediate but messages may be pushed for delivery at any point in the future.

The QueueProcessor

Polls the queue for messages and hands them off to the appropriate MessageHandler according to the message type. The QueueProcessor uses a single thread to poll the queue and invokes message handlers using a thread pool.

Using

Include com.netflix.spinnaker.keiko:keiko-redis:<version> in your build.gradle file.

Implement message types and handlers extending com.netflix.spinnaker.q.Message and com.netflix.spinnaker.q.MessageHandler respectively.

Using in a Spring Boot application

Include com.netflix.spinnaker.keiko:keiko-redis-spring:<version> in your build.gradle file. It will pull keiko-redis in transitively so there is no need to declare dependencies on both modules.

Implementing messages and handlers

For each message type you should implement a Message and a MessageHandler. A MessageHandler can also handle a heirarchy of message types.

Message implementations should be immutable value types. Kotlin data classes, Java classes annotated with Lombok's @Value, or just simple POJOs are ideal.

MessageHandler implementations may push other messages to the queue. Messages should be processed quickly. If processing time exceeds the ackTimeout value specified on the queue or an unhandled exception prevents the handler from returning normally, the message will get re-queued.

Handlers may re-push the same message they are processing in order to implement polling workflows.

Configuring the queue in Spring Boot

Telemetry

Developing a new queue implementation